Ask Hackaday: Split Rail Op Amp Power Supply

Water cooler talk at the office usually centers around movies, sports, or life events. Not at Hackaday. We have the oddest conversations and, this week, we are asking for your help. It is no secret that we have a special badge each year for Supercon. Have you ever wondered where those badges come from? Sometimes we do too. We can’t tell you what the badge is going to be for Supercon 2023, but here’s a chance for you to contribute to its design.

What I can tell you is that at least part of the badge is analog. Part, too, is digital. So we were discussing a seemingly simple question: How do we best generate a bipolar power source for the op amps on a badge? Like all design requests, this one is unreasonable. We want:

  • Ideally, we’d like a circuit to give us +/- 9 V to +/- 12 V at moderately low current, say in the tens of milliamps. Actual values TBD.
  • Low noise: analog circuitry, remember?
  • Lightweight: it is going on a badge
  • Battery operated: the badge thing again
  • Cheap: we only have a couple bucks in the budget for power
  • Available in quantity: we’ll need ~600 of these

Many of us immediately went with the “two 9V battery solution.” That’s certainly cheap and easy, but it seems heavy and fiddly for a badge. A DC-to-DC converter is probably the right way to go, but those may not be cheap or have low noise — we haven’t started down this path yet. We could do a single-sided boost and create a mid-rail virtual ground?

The Ask

We want to know your favorite trick for this! Nothing is too crazy to be out of bounds, and we know you can’t get your entire wish list, but we’d like to get as much of it as possible. We will also entertain the expensive, heavy, and noisy solutions just because we could use a good laugh. So if you were thinking of suggesting a dynamotor, go right ahead. (And take pictures!)

As long as you’re doing bipolar power design, you might as well enter it in the op amp contest, which is running right now. If you need some simulation practice with op amps, we can help there, too.

83 thoughts on “Ask Hackaday: Split Rail Op Amp Power Supply

          1. True but the max232 chip makes for a good low cost way for generating +&-12volts for analog circuits. I have seen this design pattern in many applications… as for low noise: if +-9volts is sufficient then the noise can be filtered out with some capacitors, inductors and a low noise ldo on both rails…

    1. In the old days of serial, you could steal a bit of power from the level shifter’s V+ and V- rail. Not low noise, but free ±5V if the part is there already.

      If you don’t care about accuracy, a simple voltage multiplier fed by a micro PWM output is the cheapest. With an RC filter at the end might be good enough for noise.

    2. Yeah, for the tens of milliamp range, some decent Switched Capacitor Converter with LDO’s are going to get it done and with the least space and weight possible. Since most of these come prepackaged together, minus the caps, it should be a lower BOM count to boot. If you wanna have a bunch of fun, you could make it custom and have it controlled by the “digital section.” Class G anyone?

  1. I’ve found some success using split supply generators for LCD and OLED panel biasing. Parts like TI’s TPS65131 work well, and can generate up to +/-15V at 200-300mA per rail. Also, looking at other designs like the flybuck can generate split supplies cost effectively.

  2. For DC-DC, interesting chips: R1283 (boost), LM27762 (charge pump). Depending on the PSRR of the opamps, an RC filter may be enough for cleaning up the supply rails.

    For batteries, 2x A23 is an option. They are pretty cheap in quantity, though the 50mAh capacity is pretty small.

    With modern rail-to-rail opamps, +-12V seems pretty much, and +-5V or +-3V could be worth considering.

    1. Split rail design using a sepic-cuk converter. Analog devices has a good article on this. I’ve done it with some lt sepic converter back a couple of years ago.
      Check AN-1106 from analog deivces.

    2. Hi. I’m wondering the same. I’ve done several designs with split rails, but from a 5V supply, and used a low output impedance opamp to generate a virtual ground at Vcc/2 from a resistive divider with decent filtering. Lots of opamps can operate from +/- 2.5V

      How come you need 18V to 24V rail to rail? Nowadays that’s huuuuge.

    3. We want +/- 5V on the I/O, but admittedly for compatibility/legacy “reasons”. If there’s a big hassle with getting some headroom on the power supply, I guess we could just re-spec to +/- 3.3 V (or whatever) on the I/O. That’s worth considering.

      In my experience rail-to-rail isn’t actually rail to rail. In the sense that current supply and slew rate derate toward the edges. But maybe I’m just not spending enough on the op-amps…

      1. When in doubt, recycle an old design. The C-Moy headphone amplifiers had (have?) several power options.
        Some variants attacked the issue of ground (for the amp) needing to match ground of a car. Will the proposed badge be plugging in to another device?

        1. Will be plugging into other devices, but if it’s floating on battery, that’s moot.

          Or is it? It will also have USB connectivity, so that means at least thinking about which “ground” is digital ground.

          I’ll check out the C-Moy project. Good source!

  3. If your circuitry is ok at half your supply voltage you can use a simple resistive divider feeding a single rail op amp buffer. I’ve used this trick ages ago to generate a virtual ground rail to power headphone amps off of single batteries.

  4. You probably do not need symmetric negative power supply. Just pull the whole circuit ground up some 1V using two diodes from the – pole of the battery and power the opamps directly from the battery – terminal. This way you will get some headroom for true zero input/output rail.

    1. We really do need symmetric, but the pull-up-the-circuit-ground option is a live one.

      Making the mid-rail virtual ground beefy/quiet enough is the concern. That diode trick, for instance, works great when the load is constant, but the slope of the IV curve makes the extra negative voltage padding vary with the current, right?

      1. Right, but the question is, does it matter? A decent op-amp should have something like 120+ dB supply side rejection on the output, and the circuit should be designed to not care about the rail voltages anyways because the reference should be on your virtual ground.

        If you’re doing it right, you can have all sorts of ripple and drift in your voltage rails and it won’t show in the output.

          1. One caveat: the op-amp is still inherently slow. It is intentionally limited this way to avoid instabilities, but it also makes the power supply rejection ratio dependent on frequency.


            The rejection ratio does not apply when your circuit is built in a way that the differential input voltage changes with the supply rails. That ripple will still show up at the output.

            To avoid this, you would typically add a regulator that creates a fixed voltage relative to the virtual ground. This regulator would of course have its own input rejection ratio (Example: LM317 PSRR: 65-80 dB)

          2. This is also the reason why you want to use low ESR decoupling capacitors to eat up the high frequency noise. Big electrolytics may smooth out the large scale ripple, but they don’t work for the fast transients.

  5. Coin cells, optionally rechargeable, are the simple answer. You’d only need a few of them. Capacitor-switching chips seem like the next option, or simple voltage multipliers of few stages.

    For fun wrong answers, how about using a strong led to illuminate a bunch of tiny solar cells in electrical series? (Think of the ones used for 4function calculators)

    Or you could resistively heat a couple of 12v-class thermoelectric chips although it would last an even shorter time than the first wrong answer while getting very hot. Flat, cheap, and simple though.

    Technically, a triode can produce a small voltage if you bias the grid the wrong way and let emission happen, so those could go in series too for an even worse solution.

    A little series of pager motors lined up and glued together (with no weights) might make your dynamotor idea workable.

    1. These are the right / hard / not-fully-answered questions, of course.

      BW: Audio++. Say 200 kHz?
      Noise: I dunno. The ADC / DAC combo (spoiler!) is 24 bit, so in principle we care down to the half-microvolt. In practice, I think we’ll be happy with 10 mV RMS.

  6. Inverting DC-DC plus a coupled inductor on the output. SLVA369 / TPS54040…
    Add some nice LDOs on the output.
    Nb. make sure the coupled inductor has a good coupling coefficient.

  7. As others have noted, a charge pump makes sense. Using a boost converter and some diodes and caps, you can get both rails from that (though only the positive rail is regulated)

      1. The difficulty is finding a chip that has a low active quiescent current AND high output current. That one does 2.5 mA which isn’t too bad, but then again coin cells don’t have a whole lot of capacity to spare.

  8. “Input voltages ranging from +1.5 V to +7 V can be inverted into a negative –1.5 V to –7 V output supply. This inverting scheme is ideal for generating a negative rail in single power supply systems. Only two small external capacitors are needed for the charge pump. Output currents up to 50 mA with greater than 90% efficiency are achievable, while 100 mA achieves greater than 80% efficiency.”

  9. “How do we best generate a bipolar power source for the op amps on a badge?”
    Don’t. Run your op amp circuitry with a positive rail and ground, centre your op amp signals about a fake virtual ground rail created by a potential divider between +V and Gnd (you can even buffer this with an extar op amp if you need to draw a lot of current from the virtual ground. Look in to rail-to-rail (RRIO) op amps designed for running from 5V supplies. Plenty of modern op amps are made for operating with +5V and Gnd because that’s what they’ll be given in modern USB powered devices, MCP6294 for example.

    P.S. stabilising analogue rails, if the current requirement is low then usea Pi Filter. Cap from dirty initial power rail to Gnd. Inductor from dirty power rail to feed the clean power rail. Then another cap from the clean power rail to ground. Suppresses voltage ripples very well. Larger caps and inductor let this suppression work at ever lower frequencies.

  10. LM7812s. Two of them, the first in the traditional config… the second with its output tied to the first one’s ground rail and *its* ground rail as your negative rail.

    Yes, it’s “I’ve got a jaaar of diiirt” levels of naughty circuit design ;) :D but it gets the job done. Besides, that’s the sort of devilish trickery that just makes everyone grin. We all know the line about ‘stupid’ solutions that work… I say the same axiom applies to kludge-work.

    Oh… and if you need higher current, use a pair of LM333s in an inverse form of the circuit ;)

    Bonus points if you use TO-3 packaged regulators.

    Have fun, good luck, and do try and keep the smoke INSIDE the can…

  11. Use a battery to power one DC motor. Have that drive two other motors in opposite directions to generate + and – rails. Make the motors out of PCBs for bonus points. :-Þ

  12. The _really_ quick and dirty way to do this is to make your own inverting charge pump with a couple diodes and capacitors and connect it to a microcontroller pin generating a square wave at some frequency way higher than your max analog frequency.

    Assuming your microcontroller can source/sink 20 milliamps per pin, you can get ~10ish milliamps of negative rail without needing any other ICs. Not terribly well regulated and somewhat noisy, but it works.

  13. Why focus on a single voltage of the power supply?
    If you are already going to build an SMPS, then adding an auxillary winding to the inductor to make it an transformer and a diode and capacitor may already get you there.

    Custom wound transformers may be out for only 600 pieces, but there may be “standard” transformers available that are suitable, especially if you can shift a bit with output voltages in your design.

    1. Yes Ricardo. 2 rechargeable coin cells in parallel, 2 Transistor circuit, 2 Schottky Diode and a couple of capacitors to manage the +- outputs. I’ve done it for my 9v multimeters. Just add a female socket to recharge the batteries. Smartphone battery are also a good choice. ABEGA FOUDA from Cameroon.

    2. Instead of a self-oscillationg joule thief, you can also use a uC pin with a transistor, inductor etc, to create an extra power supply voltage. When you do it with a uC, you have of course the inherent ability to turn the thing on only when needed with a bit of software. You can also add extra tricks such as turning it off for a short time when taking for example an ADC sample

  14. If the two 9 V batteries are too bulky then why not try small coin cell batteries?

    Maybe DC-DC convertors coupled with linear regulators would work well enough to reduce noise.

  15. You can use a standard boost converter to generate split rails. Take a look at the NCP1403 datasheet for an example:
    (bottom of page 16)

    Switching converters are inherently noisy though, so the output needs filtering. JFET source followers are good for that, if a little pricey: the JFET’s low transconductance means large changes in the drain-gate voltage (unregulated) will have relatively little effect on the drain current (regulated).

    Send the regulated output current through a shunt regulator like a Zener diode, and use the Zener to bias the JFET’s gate. You can also use a super-Zener like a TL431. The result is a good, quiet supply rail.

    The design isn’t great for high-current applications, but works for an op amp supply that only needs a few milliamps.

  16. Since you didnt specified any other limitation, a chain of capacitors, bang. It would probably last 1 ms, but thats the cheapest and lowest noise you can get.

    Now, where is my money?

  17. You have answered your question. As far as the choice is concerned, in case you can, use two 18650, 79A/h, Li Ion batteries with connected middle point and a TI amplifier, which works with minimum +- 2.5V. Each battery will give you 4.3V when fully charged to be used to 3.3V before the next charge.

  18. A DCDC converter is overkill. And there aren’t a lot of circuits that really need +-12V as opposed to say +-5V. There are great op amps that are very low noise and work perfectly well at +-3V. So 4 coin cells laid down flat would work great for a badge. Presumably this thing only needs to last for a few days during some conference. A PCB with clips for 4 CR2032 would be quite flat, last about 24 hours straight at 10mA and you can get a pack of 6 on Amazon for $8 USD. You can also have a half-sized two coin cell version for +-2.xV circuits (there are new CMOS op-amps that are very low noise and work well down to 2V).

  19. Admittedly, I didn’t read all comments in detail, but I’m suprised that, given no desire for two 9V batteries, no one had asked, What power do you have (or want). The solution is based on the source of power.

    My personal choice is a couple of large solar panels…

  20. 9V batteryt into 555 timer into a voltage doubler (pair of diodes and a capaciter) but use it to create a -9V rail, you can then use your positive and negative rails into some op amps to create your precision rails.

  21. Normally Op Amps are very good at rejecting any ripple on the DC power supply. However if bias voltages are developed from the power supply, or the application demands very low noise, then a very clean power source might be needed. A good design practice is to separate any reference voltages from the main power supply by using low noise reference voltage chips. If the main DC power supplies need to be cleaned up, then two voltage regulators can be used in series to provide a very smooth output. First regulate down to 15V, then a second regulator in series can regulate down to 12V. Another possibility is to use the output of an Op Amp to provide the +12 V. Set the Op Amp to produce a + or – 12 V output, and you should have some really clean power.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.